000265923 001__ 265923
000265923 005__ 20190812204801.0
000265923 037__ $$aCONF
000265923 245__ $$aRethinking Person Re-Identification with Confidence
000265923 260__ $$c2019
000265923 269__ $$a2019
000265923 300__ $$a8
000265923 336__ $$aConference Papers
000265923 520__ $$aA common challenge in person re-identification systems is to differentiate people with very similar appearances. The current learning frameworks based on cross-entropy minimization are not suited for this challenge. To tackle this issue, we propose to modify the cross-entropy loss and model confidence in the representation learning framework using three methods: label smoothing, confidence penalty, and deep variational information bottleneck. A key property of our approach is the fact that we do not make use of any hand-crafted human characteristics but rather focus our attention on the learning supervision. Although methods modeling confidence did not show significant improvements on other computer vision tasks such as object classification, we are able to show their notable effect on the task of re-identifying people outperforming state-of-the-art methods on 3 publicly available datasets. Our analysis and experiments not only offer insights into the problems that person re-id suffers from, but also provide a simple and straightforward recipe to tackle this issue.
000265923 6531_ $$aPerson Re-Identification
000265923 6531_ $$aRepresentation Learning
000265923 6531_ $$aTracking
000265923 700__ $$0254491$$aAdaimi, George$$g284287
000265923 700__ $$0261038$$aKreiss, Sven$$g296274
000265923 700__ $$0242925$$aAlahi, Alexandre$$g129343
000265923 7112_ $$aarXiv
000265923 8560_ $$fgeorge.adaimi@epfl.ch
000265923 8564_ $$zPREPRINT$$uhttps://infoscience.epfl.ch/record/265923/files/Person_Re-Identification_with_Confidence.pdf$$s3044694
000265923 909C0 $$pVITA$$malexandre.alahi@epfl.ch$$0252606$$zPasquier, Simon$$xU13529
000265923 909CO $$pconf$$pENAC$$ooai:infoscience.epfl.ch:265923
000265923 960__ $$ageorge.adaimi@epfl.ch
000265923 961__ $$afantin.reichler@epfl.ch
000265923 973__ $$aEPFL$$rREVIEWED
000265923 980__ $$aCONF
000265923 981__ $$aoverwrite